Dipesh Tharu Mahato

Dipesh Tharu Mahato, Rachel Ren

Vision Language Action models combine perception, language grounding, and control in a single policy, but their failures are hard to diagnose once visual conditions shift. We test whether OpenVLA feedforward activations contain linearly decodable information about near term task failure in LIBERO manipulation rollouts. The policy is fixed throughout. We log internal activations during execution and fit lightweight monitors after the rollouts are collected. Occlusion is the main controlled stress test. It reduces OpenVLA success from $57\%$ to $17\%$ over $100$ episodes per condition. Under this shift, a logistic probe at layer 16 reaches AUROC $0.972$ and AUPRC $0.352$ for predicting failure within a $15$ step horizon. It outperforms both a mean difference direction and an action disagreement baseline. A sparse layer sweep finds uneven decodability across depth: layer 16 is strongest among the tested layers, layer 8 remains informative, and layer 10 is weaker. To check whether the monitor is just an occlusion detector, we also evaluate color shift and camera jitter without refitting. Color shift produces no failures in this setting, so it is a benign control rather than a failure benchmark. Camera jitter does induce failures, and the occlusion trained monitor remains above random. The result is deliberately limited: OpenVLA internal states contain failure relevant structure under controlled perceptual shift, but these experiments do not establish a causal mechanism, task held out generalization, or a deployable recovery system.

Dipesh Tharu Mahato

Safety evaluations for dual-use biology assistants often measure base-model capability, refusal behavior, or jailbreak success. These metrics miss a deployment question: for a fixed base model, how does the access condition users actually see change benign utility and harmful actionable assistance? I introduce safeguard-conditioned uplift, a protocol for comparing deployed access conditions through a human-judged utility-risk frontier. I evaluate Claude Sonnet 4.6 and Gemini 3.5 Flash under helpful prompting, safety prompting, and an external safeguarded assistant on a 108-task surrogate benchmark, with the headline claim restricted to a locked 18-task held-out split. In a 600-row blinded human audit, the safeguarded assistant reduces harmful actionability relative to helpful prompting by -0.063 over 49 matched response pairs, with bootstrap 95% interval [-0.117, -0.011], while correctness changes by +0.009 with interval [-0.057, +0.077]. Adaptive, Test-B, cue-ablation, and controller-baseline checks support the measurement story but also show non-dominance: safety prompting is often strongest for Claude, while external control helps more for Gemini and can reduce benign utility. The contribution is not a universal defense. It is a deployment-level evaluation target, plus a learned risk-budgeted calibration procedure, for measuring how user-facing access conditions move the utility-risk frontier.

Dipesh Tharu Mahato, Rohan Poudel, Pramod Dhungana

Deep neural networks often achieve high accuracy, but ensuring their reliability under adversarial and distributional shifts remains a pressing challenge. We propose TriGuard, a unified safety evaluation framework that combines (1) formal robustness verification, (2) attribution entropy to quantify saliency concentration, and (3) a novel Attribution Drift Score measuring explanation stability. TriGuard reveals critical mismatches between model accuracy and interpretability: verified models can still exhibit unstable reasoning, and attribution-based signals provide complementary safety insights beyond adversarial accuracy. Extensive experiments across three datasets and five architectures show how TriGuard uncovers subtle fragilities in neural reasoning. We further demonstrate that entropy-regularized training reduces explanation drift without sacrificing performance. TriGuard advances the frontier in robust, interpretable model evaluation.